Compute orchestrationStart
Stado
GPU job queue and multi-cloud scheduler
Stado runs a fleet of GPU workers against a durable object-store queue, selects local capacity before cloud, enforces registry and quota constraints, and tracks results, failures, cost, and consumer capacity.
Product contract
- Submitters write immutable job records into the object-store queue; consumers claim by moving state into running ownership.
- Local and cloud agents publish capacity so scheduling decisions can prefer existing hardware before provisioning cloud instances.
- The coordinator evaluates priority, compatibility, registry policy, quota, zone, price, and provider availability.
- Completed output and failure evidence remain addressable by job ID after compute resources terminate.
Connected systems
- GCS
- GCP
- AWS
- Azure
- Local GPU hosts
- Box CPU sandboxes